classified-data-2026-07-06
收藏资源简介:
该数据集是一个用于机器人学研究的机器人操作数据集,由LeRobot工具创建。它记录了Agilex Piper双手机器人的操作数据,包含219个episodes,总计41,978帧数据,帧率为20 FPS。数据以分块形式存储,数据文件总大小约为100 MB,视频文件总大小约为200 MB。数据集目前仅包含训练集(所有episodes)。每个样本包括以下主要特征:1) 观测状态(observation.state):一个14维的浮点向量,表示机器人左右手臂各6个关节角度以及左右夹爪的状态;2) 动作(action):一个14维的浮点向量,表示对应的控制指令,维度与观测状态一致;3) 多视角图像观测(observation.images.*):包含来自前(front)、左(left)、右(right)、上(top)四个视角的视频数据,每个视频帧为480x640分辨率的RGB图像(3通道),编码格式为H.264;4) 图像时间戳(observation.camera_ts_ns.*):与每个视角图像对应的纳秒级时间戳;5) 索引与元数据:包括时间戳(timestamp)、帧索引(frame_index)、episode索引(episode_index)、全局索引(index)和任务索引(task_index,当前数据集仅包含1个任务)。该数据集适用于机器人模仿学习、强化学习、视觉-动作策略学习等任务的研究与模型训练。
This dataset is a robotic manipulation dataset for robotics research, created using the LeRobot tool. It records operational data from the Agilex Piper dual-arm robot, containing 219 episodes with a total of 41,978 frames at a frame rate of 20 FPS. The data is stored in chunks, with the total size of data files being approximately 100 MB and video files approximately 200 MB. The dataset currently includes only the training set (all episodes). Each sample includes the following main features: 1) Observation state (observation.state): a 14-dimensional floating-point vector representing the joint angles of the robots left and right arms (6 each) and the states of the left and right grippers; 2) Action (action): a 14-dimensional floating-point vector representing corresponding control commands, with dimensions consistent with the observation state; 3) Multi-view image observations (observation.images.*): video data from four perspectives—front, left, right, and top—with each video frame being a 480x640 resolution RGB image (3 channels) encoded in H.264 format; 4) Image timestamps (observation.camera_ts_ns.*): nanosecond-level timestamps corresponding to each perspective image; 5) Index and metadata: including timestamp, frame index, episode index, global index, and task index (the current dataset contains only one task). This dataset is suitable for research and model training in tasks such as robotic imitation learning, reinforcement learning, and vision-action policy learning.
数据集概述
- 数据集名称: classified-data-2026-07-06
- 许可证: Apache-2.0
- 任务类别: 机器人学 (Robotics)
- 标签: LeRobot
- 创建工具: LeRobot
数据集规模
- 总片段数 (Episodes): 219
- 总帧数 (Frames): 41,978
- 任务数: 1
- 帧率 (FPS): 20
- 数据文件大小: 100 MB
- 视频文件大小: 200 MB
- 分块大小: 1000
数据集划分
- 训练集 (Train): 片段索引 0 至 218 (共219个片段)
机器人类型
- 机器人型号: agilex_piper_bimanual (双臂机器人)
特征信息
| 特征名称 | 数据类型 | 形状 | 描述 |
|---|---|---|---|
observation.state |
float32 | (14,) | 包含14个维度:右臂6个关节、左臂6个关节、右夹爪、左夹爪 |
action |
float32 | (14,) | 动作空间,与状态维度对应(右臂6关节、左臂6关节、右夹爪、左夹爪) |
observation.images.front |
视频 | (480, 640, 3) | 前向相机图像,H.264编码,20 FPS,分辨率480x640 |
observation.images.left |
视频 | (480, 640, 3) | 左侧相机图像,编码与分辨率同上 |
observation.images.right |
视频 | (480, 640, 3) | 右侧相机图像,编码与分辨率同上 |
observation.images.top |
视频 | (480, 640, 3) | 顶部相机图像,编码与分辨率同上 |
observation.camera_ts_ns.front.rgb |
int64 | (1,) | 前向相机时间戳(纳秒) |
observation.camera_ts_ns.left.rgb |
int64 | (1,) | 左侧相机时间戳(纳秒) |
observation.camera_ts_ns.right.rgb |
int64 | (1,) | 右侧相机时间戳(纳秒) |
observation.camera_ts_ns.top.rgb |
int64 | (1,) | 顶部相机时间戳(纳秒) |
timestamp |
float32 | (1,) | 时间戳 |
frame_index |
int64 | (1,) | 帧索引 |
episode_index |
int64 | (1,) | 片段索引 |
index |
int64 | (1,) | 全局索引 |
task_index |
int64 | (1,) | 任务索引 |
数据格式
- 数据文件: Parquet 格式,路径为
data/chunk-{chunk_index:03d}/file-{file_index:03d}.parquet - 视频文件: MP4 (H.264编码),路径为
videos/{video_key}/chunk-{chunk_index:03d}/file-{file_index:03d}.mp4 - 代码库版本: v3.0



